mlfix

mlflow artifact store migration fix tool

Motivation

Currently the is no way to natively make models trained using mlflow portable across local machines or from remote to local, while keeping all the benefits (working commands etc.) of the mlflow environment (see for example this issue).

This is especially bad in small teams, local development or just prototyping.

This tool makes it easy to fix existing mlflow artifact store to current path.

Future work may include migrating existing artifact stores, only specific experiments etc.

Currently mlf-core also supports such functionality, but if you are not using mlf-core and want just to fix your mlruns, this tiny tool will help you.

Installation

This is tested for Python 3.6 to 3.9.

From PyPI:

$ pip install mlfix

From the source code (in the main directory):

$ python -m pip install .

Usage

$ mlfix path_to_artifact_store

That is it!

You must specify the name of the mlruns folder if it was different than the default in the former location of the store:

$ mlfix --mlruns-name nonstandard_mlruns path_to_artifact_store